generation_config.json

{
  "bos_token_id": 2,
  "do_sample": true,
  "eos_token_id": [
    1,
    1,
    106
  ],
  "max_length": 4096,
  "max_new_tokens": 1024,
  "no_repeat_ngram_size": 3,
  "pad_token_id": 0,
  "repetition_penalty": 1.35,
  "stop_sequences": [
    "</think>",
    "</reasoning_step>"
  ],
  "temperature": 0.4,
  "top_k": 40,
  "top_p": 0.95,
  "transformers_version": "4.57.1"
}